Measure whether you're actually getting cited

Baseline your share of voice

You can't improve what you don't measure, and traditional rank trackers don't see AI answers. Build a fixed set of the prompts your buyers actually ask, run them across the engines, and record whether — and how — you're mentioned. That baseline is your scoreboard. Because citations churn 40–60% week to week — a University of St. Gallen study measured about 65% of cited sources changing from one day to the next — track the trend, never a single run, and always score per platform rather than blending engines into one number: overlap between them is tiny.

Help me measure my AI share of voice.
1. From my topic/products (below), generate 25 prompts real buyers would type into ChatGPT,
   Perplexity, Gemini, and AI Mode — a mix of category, comparison, and problem queries.
2. I'll run them and paste the answers back. For each answer, score: is my brand mentioned?
   linked? cited as a source? which competitors appear? what sources did the engine cite?
3. Summarize my visibility %, share of voice vs named competitors, and the 5 sources most often
   cited that I don't yet appear on — those become off-page targets.

Catch AI referrals hiding in your analytics

AI traffic under-reports. Many AI referrals land as Direct or (none), and name-only tracking misses "ghost citations" where an engine cites your URL without naming your brand. Build a referrer channel group for the AI sources and add a self-reported "how did you hear about us" field to triangulate.

Score with one rubric, per platform

Free-form notes rot. Score every answer on the same four-point scale, log the same columns, and keep engines separate — with ~11% overlap between platforms, a blended number is fiction.

Alongside the score, log sentiment, an accuracy flag (misquotes, wrong prices), competitors named, and sources cited. Average two or three runs per checkpoint — daily churn is ~65% — and chart each platform on its own line.

Here are this week's answers for my fixed prompt set, labeled by platform and prompt. Build a table: platform | prompt | score 0-3 (0 absent, 1 mentioned, 2 cited/linked, 3 recommended first) | sentiment | accuracy issues | competitors named | sources cited. Then compare against the baseline table I'm pasting: the three biggest movers up and down, and any factual errors about my brand I should correct at the source.

Referral numbers are a floor — find the ghost zone

In GA4, a custom channel group catching referrer sources that match chatgpt|openai|perplexity|copilot|gemini|claude|poe captures the visible slice of AI traffic — typically around 1% of sessions, converting several multiples better. Treat it as a floor, not a census: assistant apps often strip referrers, so real usage lands in Direct.

The invisible slice is measurable from the other side. Pull thirty days of server logs, filter for OAI-SearchBot, ChatGPT-User and PerplexityBot hits on your money pages, and compare against referral sessions to those same pages. A big gap means your content is feeding answers without attribution clicks — the 40–60% ghost zone — which is an argument for stronger in-answer branding (named data, named frameworks), not despair.

Set a measurement cadence

1. Weekly: re-run your top-priority prompts; note movement, expect churn.
2. Monthly: full query set + GA4 AI-channel review + crawler-log check.
3. Quarterly: re-verify every stat and crawler name, and re-audit (back to Phase 0).